Estimating the Price Elasticity of Ferrous Scrap Supply

Transcription

1 Estimating the Price Elasticity of Ferrous Scrap Supply Robert J. Damuth Economist and Principal Consultant Nathan Associates Inc. February 25, 2011

2

3 Contents 1. Introduction 1 2. Background Relevance of Elasticity Why the Need to Model Demand and Supply? A Basic Model Estimating the Model Logarithmic Transformation for Estimating Elasticity Data Generating a Series on Purchased Obsolete Ferrous Scrap Start with Total Purchased Scrap Receipts Estimate Prompt Scrap Generated Derive Purchased Obsolete Scrap Scrap Prices Factors Affecting Scrap Supply Other than Price Scrap Processing Cost Scrap Transportation Cost Apparent Steel Consumption Ferrous Scrap Trade Economic Shocks Factors Affecting Scrap Demand Other than Price Consumption of Iron Ore Crude Steel Production Technological Change Availability of Other Scrap 20

4 II 4. Model Total Prompt and Obsolete Scrap Obsolete Scrap Estimation Results Total Prompt and Obsolete Scrap Supply Obsolete Scrap Supply 26 Appendix A Appendix B ILLUSTRATIONS Figures 1. Price, Net Additions to Inventory, and Recovered Scrap, Price Sensitivity of Demand and Supply as Measured by Elasticities 5 3. Effects of Supply Elasticity on Market-Clearing Prices and Quantities 6 4. Price and Quantity Data Are Transaction Amounts 8 5. The Identification Problem 8 Tables 1. Correlation between Price and Scrap Recovered 4 2. Regression Results for Estimation of Price Proxy (first stage of 2SLS) Regression Results for Supply of Total Prompt and Obsolete Ferrous Scrap Regression Results for Supply of Obsolete Ferrous Scrap 27

5 1. Introduction Nathan Associates Inc. analyzed the relationship between ferrous scrap prices and the quantities of obsolete ferrous scrap recovered from the national inventory and supplied to the U.S. market for scrap. Nathan assembled a database of quarterly observations on scrap prices and quantities transacted from 1985 through 2009, as well as on factors other than price that affect the supply of and demand for scrap. Simultaneous equation models of scrap supply and demand in log-linear form were specified and estimated using two-stage least squares. Nathan found that since their earlier study of 1979, the sensitivity of obsolete ferrous scrap supply to changes in scrap prices has increased. A 10 percent increase in scrap prices elicits an 8.85 percent increase in the quantity of obsolete scrap supplied to the U.S. market. Although steel producers sometimes call for limits on scrap exports based on claims that domestic scrap supplies are inadequate, the Nathan study in combination with the firm s studies of the growing national inventory of obsolete ferrous scrap reveals that additional supplies of obsolete ferrous scrap will be recovered from the national inventory and supplied to the U.S. market as scrap prices increase. Building on a long history of estimating the national inventory of obsolete ferrous scrap, 1 Nathan Associates Inc. recently completed an update of its first study 2 of the price elasticity of ferrous scrap supply. This report presents the update. It focuses on the relationship between changes in the price of ferrous scrap and the amount of obsolete ferrous scrap recovered from the national inventory and supplied to the market. Steel manufactures sometimes claim the supply of scrap is insufficient for their needs. Hence, they call for limitations on exports of scrap. However, Nathan Associates has demonstrated in its past studies of scrap inventories that, in fact, the national inventory of obsolete ferrous scrap has continued to build as more and more ferrous containing end-use products reached the ends of their useful lives and were discarded. In our most recent update, the national inventory as of December 31, 2009 was 1.18 billion tons, 1.04 million tons of which accumulated from 2004 through During this period, discarded end-use products contained an average of 87.2 million tons of ferrous 1 Obsolete ferrous scrap is ferrous material that exists in discarded end-use products such as automobiles, household appliances, construction materials, and others. For the most recent Nathan Associates Inc. study of the national inventory of obsolete ferrous scrap, see Iron and Steel Scrap: Accumulation and Availability as of December 31, 2009, Institute of Scrap Recycling Industries, Inc. (ISRI), Washington, DC, See Robert R. Nathan Associates, Inc., Price-Volume Relationships for the Supply of Scrap Iron and Steel: A Study of the Price Elasticity of Supply, Scrap Metal Research and Education Foundation, Washington, DC, Jan. 8, 1979.

6 2 material of which 65 million tons were recoverable. Forty-seven and one-half million tons were recovered annually, leaving 17.5 million tons of recoverable but unrecovered scrap each year. After accounting for annual corrosion losses, million net tons were added each year to the national inventory of obsolete ferrous scrap. Definition of National Inventory (Potential Reserves) of Obsolete Ferrous Scrap Material that is of known or inferred quantity in a condition that allows for immediate use and which is available for recovery within the constraints of known technology and higher but realistic prices. Hence, the issue is not availability, but instead how much higher the price of scrap must be to incentivize the recovery of additional scrap. While all manufactures seek to minimize their cost of materials, the economics of supply and demand combined with the recyclable nature of ferrous scrap ensure that additional scrap can be obtained at prices higher than manufactures currently pay. The question is how much higher. Our 1979 study was the first to estimate the price elasticity of obsolete ferrous scrap supply. It examined the period 1961 through Quarterly data were used to estimate average quarterly elasticities, as well as the average elasticity over the entire period. The average price elasticity of obsolete scrap supply over the entire period was Total scrap supply, which includes prompt industrial scrap 3 and obsolete scrap, was less sensitive to a change in price. Its elasticity was Our current study differs from the original in three substantive ways. First, we developed a newer database covering 1985 through In this more recent timeframe, steelmaking relied more heavily on the electric arc furnace (EAF) process, a process in which the metallic charge is 100 percent scrap. 4 In 1960, EAF steel production totaled approximately 8.4 million tons. By 1995, it totaled 42.4 million tons or 39 percent of total steel production. 5 Hence, the market for scrap in our period of analysis was different from the market that existed during the period of our original study. The second difference pertains to factors included in our analysis that explain demand for and supply of ferrous scrap. Most notably, we used apparent steel consumption instead of foundry and mill shipments of iron and steel as a factor affecting the supply of purchased total (prompt plus obsolete) scrap. 3 Prompt industrial ferrous scrap is a by-product of industrial manufacturing. It consists of items such as punchings, stampings, turnings, and borings. 4 The steelmaking industry includes EAF producers or minimills and integrated steel producers which rely on the basic oxygen furnace (BOF) technology. The metallic charge to the BOF consists of 60 percent to 70 percent hot metal from the blast furnace and 20 percent to 40 percent steel scrap. See Commentary: Introduction to Electric Arc Furnace Steelmaking, The Electric Power Research Institute, Inc. (EPRI), Center for Materials Production (CMP), 1997, which is available at 5 See Commentary: Introduction to Electric Arc Furnace Steelmaking.

7 3 Finally, in this update we specified and estimated models in log-linear form. Previously, they were not. Hence, average elasticities had to be calculated for changes in average prices and quantities. Log-linear models allow estimation of elasticities that do not vary with price and quantity. Consequently, they are often referred to as constant elasticity models. From 1991 through 2009, prices and quantities of ferrous scrap moved similarly (Figure 1). When price increased, more obsolete scrap was recovered and recycled. 6 From 1991 through 2009, the Pearson correlation coefficient 7 between price and recovered obsolete scrap was 0.86 (Table 1), which indicates a significant degree of positive correlation. The price of scrap and inventory net additions recoverable obsolete scrap less obsolete scrap recovered moved inversely. When the price of scrap increased, net additions to inventory decreased. This inverse relationship is to be expected given the positive correlation between the amount of obsolete scrap recovered and changes in the price of scrap. The Pearson correlation coefficient between price and net inventory additions was However, a correlation coefficient is not an elasticity. To measure elasticity, more complex statistical analysis is required. Here, we present an econometric analysis of scrap supply and present estimates of price elasticity. Our report is organized into four chapters. Following this introduction, Chapter 2 presents a discussion of the basic economics of supply elasticity and econometric modeling of supply and demand. Chapter 3 discusses the data requirements of our study and identifies sources we relied on to construct our database. Chapter 4 presents and discusses the specifications of models we estimated. Chapter 5 presents our results. The database is in Appendix A. Appendix B presents additional results of our analysis of scrap demand. 6 Note the lack of co-movement between price and the quantity of recovered prompt scrap, which is to be expected given that prompt scrap is a waste product of manufacturing end-use products. It is relatively easy to recover by the need to simply clear the shop floor. Hence, the availability of prompt scrap is largely unconnected to ferrous scrap prices. 7 The Pearson correlation coefficient is the standard statistical measurement of linear dependence (correlation) between two variables. Its value ranges from -1 to +1. Values closer to either -1 (indicating negative correlation) or +1 (indicating positive correlation) indicate stronger correlation.

8 4 Figure 1 Price, Net Additions to Inventory, and Recovered Scrap, , , Thousand net tons 50,000 40,000 30,000 20, $ per net ton 10, Recovered obsolete ferrous scrap Recovered prompt ferrous scrap Net additions to potential reserves Price ($ per net ton) SOURCES: Price ($ per net ton) is calculated from the price ($ per metric ton) series reported by the U.S. Geological Survey (USGS) in its annual publication Minerals Commodity Summary for Iron and Steel Scrap, , available online at See Appendix A, Table A-2 of the recent Nathan Associates Inc. study of the national inventory of obsolete ferrous scrap, Iron and Steel Scrap: Accumulation and Availability as of December 31, 2009, Institute of Scrap Recycling Industries, Inc. (ISRI), Washington, DC, 2010 for data on recovered obsolete scrap, recoverable prompt scrap, and net additions to potential reserves. Table 1 Correlation between Price and Scrap Recovered Scrap Quantity Pearson Correlation Coefficient Recovered Obsolete scrap* Recovered Prompt scrap Net Additions to potential reserves* NOTE: * indicates the correlation is statistically significant at the one percent level which implies a probability of one percent or less that the correlation occurred by chance. SOURCE: Nathan Associates Inc.

9 2. Background Price elasticity of demand or supply is simply a way of characterizing the sensitivity of product quantity demanded or supplied to a change in the price of the product. It is measured by the ratio of percentage change in quantity (demanded or supplied) to percentage change in price. If the percentage change in quantity exceeds the percentage change in price, demand or supply is more sensitive to a change in price. If the percentage change in quantity is less than the percentage change in price, demand or supply is less sensitive to a change in price. Figure 2 illustrates the basic idea of demand and supply elasticity. Generally speaking, the steeper the demand or supply curve, the less sensitive it is to a change in price it is less elastic. The flatter the curve, the more sensitive it is to a change in price it is more elastic. Figure 2 Price Sensitivity of Demand and Supply as Measured by Elasticities Demand Supply 0 < Elasticity < 1 Elasticity = 1 Price. Elasticity > 1 Price. Elasticity > 1 Elasticity = 1 Elasticity < 1 Quantity Quantity 2.1 Relevance of Elasticity Before proceeding, it is useful to consider how supply elasticity affects market-clearing prices and quantities (Figure 3). Total ferrous scrap supply (the green line of Figure 3) consists of home and prompt scrap (the yellow line), as well as obsolete scrap (the blue line). Home scrap is reprocessed in the steel mill in which it was produced. It is not supplied to the scrap

10 6 Figure 3 Effects of Supply Elasticity on Market-Clearing Prices and Quantities Price per ton Supply Home & prompt Less Elastic Supply Supply Obsolete Supply Total Demand Total original Demand Total new P** P*. Q* Obsolete Q Home & prompt Q* Total.. Quantity in tons Q** Obsolete Q** Total Price per ton Supply Home & prompt More Elastic Supply Demand Total original Demand Total new Supply Obsolete Supply Total P P. Q Home & prompt.. Q Total Quantity in tons Q Obsolete Q Obsolete Q Total

11 7 market. Prompt scrap is supplied to the market, but amounts supplied do not vary significantly, if at all, with scrap prices. Instead, prompt industrial scrap supplied to the market varies with manufacturing activity. As manufacturing increases, additional prompt scrap is generated, removed from the shop floor, and supplied to the market. On the other hand, the quantity of obsolete scrap supplied does vary with scrap prices. As prices increase, additional obsolete scrap will be recovered and supplied. Consider first the effects when demand remains constant. Market clearing prices and quantities are determined when demand equals supply, that is, where the red solid line representing demand in Figure 3 intersects the green line representing total ferrous scrap supply. When supply is more elastic, as depicted in the bottom panel of Figure 3, the total supply curve is flatter and the market-clearing quantity of scrap is greater (Q Total > Q* Total ) and the market clearing price of scrap is lower (P < P*). Now consider a shift in demand. Assume a new minimill opens. As a result, at every price point, the quantity of scrap demanded will increase. The demand curve shifts to the right, as illustrated in the top and bottom panels of Figure 3 by the dashed red line. Not surprisingly, with a more elastic supply of obsolete ferrous scrap, the increase in demand for scrap has a greater effect on quantity supplied and price. The quantity supplied is greater (Q Total > Q** Total ) and the price paid is lower (P < P**) when supply is more elastic. With scrap supply more sensitive to a change in scrap price, an increase in demand for scrap will result in additional quantities of obsolete ferrous scrap being recovered from the national inventory and supplied to the scrap market. 2.2 Why the Need to Model Demand and Supply? The previous discussion reveals a conundrum. Prices and quantities available for calculating a supply or demand elasticity are transaction data. We do not observe market demand or supply schedules, that is, what a consumer would be willing to pay or the price a supplier would be willing to accept. We observe only the prices and quantities transacted the points of intersection between demand and supply curves (Figure 4). For example, in the top panel of Figure 3, P* and Q* Total would be reported, but additional price-quantity combinations along the total supply curve would not.

12 8 Figure 4 Price and Quantity Data Are Transaction Amounts Price Price Demand. 1 Supply.. 2 Demand Supply Supply 3 Supply. Demand 4 Demand Quantity Quantity Compounding this conundrum is the fact that quantity demanded, quantity supplied, and price are determined simultaneously, which might lead to an identification problem (Figure 5). When we observe price-quantity combinations, we might be observing movement along a supply curve or movement along a demand curve. For example, as one s income increases, one s consumption will likely increase, that is, demand will increase. The demand curve will shift to the right. With a stable supply curve, points of intersection between demand and supply curves would trace points along the supply curve. On the other hand, consider a market in which demand is stable but supply is increasing, perhaps from technological innovation that allows additional quantities supplied to the market at each price point. In this case, the supply curve shifts to the right and points of intersection between demand and supply would trace points along the stable demand curve. Figure 5 The Identification Problem Stable Supply with Increasing Demand Stable Demand with Increasing Supply Price..... Supply Price Supply Demand Demand Quantity Quantity Because data reported are transaction data; the determination of price, quantity supplied, and quantity demanded occur simultaneously; and factors other than price affect supply and demand we cannot calculate elasticity as simply the ratio of a reported percentage change in

13 9 quantity to a reported percentage change in price. Instead, we must specify and estimate a model of demand and supply, including factors affecting quantities supplied and demanded other than price. 2.3 A Basic Model The simplest model of demand and supply consists of the following two simultaneous equations plus the identity or equilibrium condition: 8 1. Demand function: Q d t = A 1 + A 2 P t + u 1t 2. Supply function: Q S t = B 1 + B 2 P t + u 2t. Equilibrium condition: Q d t = Q S t where Q d t is quantity demanded at time t, Q S t is quantity supplied at time t, P t is price at time t, u 1t is the effect on demand of factors other than price, such as income, and u 2t is the effect on supply of factors other than price, such as weather. The A s and B s are structural parameters. We expect A 2 to be negative (a negative correlation between price and quantity demanded which results in a downward sloping demand curve) and B 2 to be positive (a positive correlation between price and quantity supplied which results in an upward sloping supply curve). The equilibrium condition can be re-written in the parameters of the model and solved for P t as follows: 3. A 1 + A 2 P t + u 1t = B 1 + B 2 P t + u 2t 4. P t = π 1 + v 1t where π 1 = (B 1 A 1 ) (A 2 B 2 ) and v 1t = (u 2t u 1t ) x (A 2 B 2 ). We can now substitute the P t equation into either the demand or supply equation (equation 1 or 2, respectively) to solve for Q t as follows: 5. Q t = π 2 + v 2t where π 2 = (A 2 B 1 A 1 B 2 ) (A 2 B 2 ) and v 2t = (A 2 u 2t B 2 u 1t ) (A 2 B 2 ). This simple model illustrates the simultaneity and identification problems discussed earlier. Regarding simultaneity, notice from equation 1 that if u 1t is positive, Q d t will be greater at every P t, that is, the demand curve will shift to the right. Once this happens the intersection of the new demand curve with the unchanged supply curve will occur at a new P t which then feeds back into the supply curve of equation 2. A shift in either u 1t or u 2t changes P t and Q t. Price and quantity are said to be jointly dependent. Regarding the identification problem, 8 The discussion in the section closely follows Damodar N. Gujarti s Chapter 15, Simultaneous Equation Models, in Essentials of Econometrics 3 rd ed., McGraw-Hill Irwin, 2006.

14 10 notice that equations 4 and 5, which are referred to as the reduced form of the model, include the four structural parameters (the A s and B s) of the model. However, one cannot estimate four parameters in a model of only two equations. One needs at least four independent equations to estimate four parameters. In this model, demand and supply are under-identified. 2.4 Estimating the Model Estimating a model of simultaneous equations first requires determining whether the model s specification is under-identified, exactly identified, or over-identified. If the model is underidentified, such as our simplest model, no statistical technique can be used to estimate the model s structural parameters. If the model is exactly identified, the method of indirect least squares can be used. If the model is over-identified, the method of two-stage least squares (2SLS) can be applied. Before turning to a description of 2SLS, which is the technique we applied to estimate the parameters of our ferrous scrap models, consider how one can determine whether a model is under-, exactly, or over-identified. The simplest way is to count the number of endogenously determined variables, which is equivalent to the number of equations in the model, and the number of exogenously determined variables (variables whose values are determined outside the model). In our simplest model, there are two endogenous variables and no exogenous variables. Although the supply and demand equations in the simplest model include terms that capture the total effect of factors other than price on demand (the u 1t term) and supply (the u 2t term), these terms are stochastic (random) error terms. The rules in making a determination of identification are as follows: If the total (k) number of variables (both endogenous and exogenous) that are excluded from the model s equation under consideration is less than the number of endogenous variables in the model minus one (m 1), the equation is under-identified, that is, the equation is under-identified if k < m 1. If the total (k) number of variables (both endogenous and exogenous) that are excluded from the model s equation under consideration equals the number of endogenous variables in the model minus one (m 1), the equation is exactly identified, that is, the equation is exactly identified if k = m - 1. If the total (k) number of variables (both endogenous and exogenous) that are excluded from the model s equation under consideration is greater than the number of endogenous variables in the model minus one (m 1), the equation is over-identified, that is, the equation is over-identified if k > m 1.

15 11 Returning now to our simplest model, in both the demand and supply equations (equations 1 and 2, respectively), there are no exogenously determined variables so none is excluded from either (k =0). Hence, for both supply and demand equations, k is less than (m 1), that is, 0 < 1, and both are revealed to be under-identified, a result that was previously determined by deriving the reduced form of the model and realizing its two equations contained four structural parameters. Now consider modifying the simplest model by including income in time t (I t ) as an exogenous factor in the determination of demand. This augmented simple model can be written as 6. Demand function: Q d t = A 1 + A 2 P t + A 3 I t + u 1t 7. Supply function: Q S t = B 1 + B 2 P t + u 2t. Here, m remains equal to two, but k equals zero in the demand function and one in the supply function. Hence, in our augmented model the demand function is under-identified (k < m - 1) while the supply function is exactly identified (k = m 1). By adding income to the demand function, the supply function is identified. This result was illustrated in the left panel of Figure 5 above. Recall how a shifting demand curve (caused by increasing income) with a stable supply curve allowed identification of price-quantity combinations along the supply curve. As a final example, consider two more modifications to our simplest model. In addition to adding income to the demand function, we add a binary variable (S t ) that equals one during summer months and zero otherwise. Such a factor is often included to account for seasonal differences in demand. We also add price in the period prior to t to the supply function. This revised model can be written as 8. Demand function: Q d t = A 1 + A 2 P t + A 3 I t + A 4 S t + u 1t 9. Supply function: Q S t = B 1 + B 2 P t + B 3 P t 1 + u 2t. In this revised model, m remains equal to two. However, k in the demand function now equals one because P t 1 (which is an exogenous variable because its value is known at time t) is excluded from the demand function. The value of k in the supply function equals two because it excludes I t and S t. Hence, the demand equation is exactly identified (k = m 1) and the supply equation is over-identified (k > m 1). When an equation is over-identified, 2SLS can be used to estimate its structural parameters. 2SLS applies the standard statistical method of ordinary least squares (OLS) 9 twice. First, a proxy for P t that is not correlated with the error term u 2t of the supply curve must be specified 9 See Gujarati, Chapter 6 for an introduction to OLS.

16 12 and estimated. One such proxy is P t itself as a function of all exogenous variables in the model, such as: 10. P t = C 1 + C 2 I t + C 3 S t + C 4 P t 1 + w t where w t is a random error term. After estimating P t, one substitutes the estimated values into the supply equation and estimates its parameters using OLS once again. In other words, the supply function is estimated by regressing Q S t on estimated P t s, as follows, not on the original P t s. 11. Q S t = B 1 + B 2 (EstimatedP t ) + B 3 P t 1 + e t where e t is a random error term. 2.5 Logarithmic Transformation for Estimating Elasticity Recall that elasticity is defined as the percentage change in quantity divided by the percentage change in price. Elasticity can be written as follows: [(Q New Q Old ) Q Old ] [(P New P Old ) P Old ], which can be rewritten as (dq/q) (dp/p), and finally as (dq/dp) (P/Q). In the supply function of the simplest model, quantity supplied was specified as a linear function of price as follows: Q S t = B 1 + B 2 P t + u 2t In this linear function, dq S t/dp t equals B 2 and P t /Q S t = P t /(B 1 + B 2 P t + u 2t ). Hence, the supply elasticity can be rewritten as B 2 [P t /(B 1 + B 2 P t + u 2t ] and one sees that supply elasticity varies with price. Even with the parameters (B 1 and B 2 ) of the function estimated, one cannot calculate elasticity until a price is given, and the calculated value can be different for different prices.

17 13 However, a logarithmic transformation of the supply function enables calculation of a constant elasticity. Rewrite the supply function by taking the natural logarithm (ln) of both sides of equation 2 above as follows: 12. ln(q S t) = B 1 + B 2 ln(p t ) + u 2t Recognizing a change in the logarithm of a number is a relative or proportional change, that is, d(lnq S t) equals dq S t/q S t and d(lnp t ) equals dp t /P t, we see that B 2, which equals d(ln(q S t))/d(ln(p t )), is the percentage change in quantity (dq S t/q S t) divided by the percentage change in price (dp t /P t ). The estimated price parameter (B 2 in equation 12 above) is the price elasticity of supply, and, more important, it is constant and does not vary as price varies. By transforming the function into a log-linear form, we can estimate a constant price elasticity. Having introduced these basic concepts, we next turn to a discussion of the data required to model ferrous scrap supply and demand.

18

19 3. Data Our study focuses on the responsiveness of ferrous scrap supply to changes in price. Hence, a database of amounts purchased, prices, factors affecting scrap supply, and factors affecting scrap demand must be assembled. Our database consists of quarterly observations for 1985 through It is presented in Appendix A. 3.1 Generating a Series on Purchased Obsolete Ferrous Scrap We must distinguish between obsolete and prompt scrap to determine the amount of obsolete scrap recovered from the national inventory. However, statistics on purchased scrap are available for total prompt and obsolete. Reported data do not separate purchased prompt from purchased obsolete scrap. Hence, we begin by deriving a historical series on purchased amounts of obsolete ferrous scrap. We do so by relying heavily on our most recent study of the national inventory of obsolete ferrous scrap START WITH TOTAL PURCHASED SCRAP RECEIPTS Annual amounts of total purchased scrap are reported in the U.S. Geological Survey (USGS) of the U.S. Department of the Interior. Net receipts include receipts of scrap from brokers, dealers, and other outside sources, plus receipts from other own-company plants, minus shipments. Receipts are reported in thousand metric tons. The series can be constructed by subtracting shipments from receipts for 1985 through 1990 as reported in USGS Table 1: Salient U.S. Iron and Steel Scrap and Pig Iron Statistics. 11 For years after 1990, the series is reported as net receipts of ferrous scrap for all manufacturing types in Table 1: Salient U.S. Iron and Steel Scrap, Pig Iron, and Direct-Reduced Iron Statistics of the Minerals Yearbook. 12 We converted annual net receipts into quarterly amounts using U.S. quarterly total crude steel 10 See Nathan Associates Inc., Iron and Steel Scrap, Accumulation and Availability as of December 31, 2009, Institute of Scrap Recycling Industries, Washington, DC, 2010 for additional details on the derivation of obsolete scrap consumption. 11 See 12 The Minerals Yearbook is an annual publication of USGS. It is available as portable document format (pdf) files at

20 16 production reported by the World Steel Association (formerly the International Iron and Steel Institute) ESTIMATE PROMPT SCRAP GENERATED We applied the following three-step process to estimate quantities of prompt scrap generated: First we calculated annual total amounts of ferrous material consumed by manufacturers of end-use products. 14 Next we disaggregated annual total amounts into 19 iron and steel consuming sectors, 15 such as construction materials, consumer durables, and automotive, among others. Then we multiplied sector-specific prompt scrap generation rates, that is, prompt scrap generated during the manufacture of ferrous-containing end-use products, by the quantities of ferrous material consumed by each sector. Rates are based on a study by Hogan and Kolke. 16 They range from six percent to 31 percent, with an average across all 19 ferrous consuming sectors of 16.6 percent DERIVE PURCHASED OBSOLETE SCRAP Purchased obsolete scrap is purchased total scrap less scrap imports plus scrap exports less prompt scrap generated. Prompt scrap generated is reduced 1.5 percent to account for waste before subtracting it from trade adjusted purchased total scrap. 18 Imports and exports are reported by USGS at Scrap Prices For scrap prices, we used the producer price index (PPI) for iron and steel scrap published by the Bureau of Labor Statistics (BLS) of the U.S. Department of Labor. 19 It measures average monthly change in prices received by domestic producers of iron and steel scrap. To derive a quarterly series, we averaged monthly values across three-month periods. 13 See 14 Derived from various sources such as the American Iron and Steel Institute, the Census of World Casting published by Modern Casting, and Current Industrial Reports published by the U.S. Census Bureau. 15 As defined by the American Iron and Steel Institute. 16 William T. Hogan and Frank T. Koelble, Purchased Ferrous Scrap Demand and Supply Outlook, Industrial Economic Research Institute, Fordham University, New York, June Based on personal conversations with researchers, the Steel Recycling Institute, USGS, and presentations given at the International Iron and Steel Institute, we found that the results of the 1977 study by Hogan and Kolbe remain accurate. 18 The waste percentage figure, or a prompt recovery rate of 98.5 percent, is based on an internal study conducted by the Steel Recycling Institute in This information was provided in a telephone conversation with the Steel Recycling Institute on 1/18/ The PPI for iron and steel scrap is reported as series WPU1012 at ftp://ftp.bls.gov/pub/time.series/wp/wp.data.11a.metals Additional information is available at

21 Factors Affecting Scrap Supply Other than Price Factors other than scrap price that are likely to affect scrap supply include the cost of processing scrap, transportation cost, fluctuations in steel mill and foundry shipments, scrap trade, and events that can shock an economy into abnormal performance SCRAP PROCESSING COST Data on scrap processing cost per se are not available. However, the major drivers of processing cost include labor cost, energy cost, and cost of raw materials. Hence, we constructed a cost index as a proxy for scrap processing cost and included it in our supply function. The index is a weighted composite of the three major drivers, with weights of 10.5 percent for labor cost, 7.0 percent for energy cost, and 82.5 percent for raw material cost. 20 For labor cost, we used BLS-reported private sector average hourly earnings of production and nonsupervisory employees. 21 For energy cost, we used the BLS producer price index (not seasonally adjusted) for fuels and related products and power. 22 For raw materials, we used the BLS producer price index (not seasonally adjusted) for industrial commodities SCRAP TRANSPORTATION COST Because prices of ferrous scrap are quoted at the point of delivery, transportation cost is an important factor affecting scrap supply. As a proxy for transportation cost, we included railroad revenue per ton-mile of transported waste and scrap in our supply function. It is reported by the Surface Transportation Board (STB) of the U.S. Department of Transportation (DOT). 24 Data are annual through We estimated values for 2008 and 2009 using the historical compound annual growth rate from 20 Weights provided by a large processor of ferrous scrap. 21 See series CES , average hourly earnings of production and nonsupervisory employees, at =all_periods&output_format=text&reformat=true&request_action=get_data&initial_request=false&data_tool =surveymost&output_type=column&series_id=ces See series WPU05, fuels and related products and power, at =all_periods&output_format=text&reformat=true&request_action=get_data&initial_request=false&data_tool =surveymost&output_type=column&series_id=wpu See series WPU03THRU15, industrial commodities, at =all_periods&output_format=text&reformat=true&request_action=get_data&initial_request=false&data_tool =surveymost&output_type=column&series_id=wpu03thru Revenue per ton-mile can be obtained by downloading a 2.94MB Excel workbook at Click on Click here to view Rate Study workbook and save the file. The sheet titled Summary_Statistics contains waste and scrap (commodity code 40) revenue per ton-mile (RPTM) for 1985 through Each year has an entry for short, medium, and long haul trips. We combined revenue and ton-miles reported for each distance category and divided total revenue across all three distance categories by total ton-miles across all three distance categories to calculate RPTM across all three categories in each year.

22 through We made no attempt to estimate a quarterly series. Instead, we used the annual statistic for each quarter of the year APPARENT STEEL CONSUMPTION Prompt scrap is generated when steel products (sheets, plates, bars, and wire, among others) are consumed in the production of end-use products. 25 Hence, the consumption of steel products is likely to affect the supply of total purchased scrap. We obtained a historical series of apparent steel consumptions, which equals production minus exports plus imports, from the USGS FERROUS SCRAP TRADE To control for the impact of U.S. exports of scrap on U.S. supply, we include the ratio of scrap exports 27 to domestic purchased scrap receipts in our supply function. USGS reports scrap exports and purchased scrap receipts on an annual basis. Again we used quarterly crude steel production data to convert annual scrap export ratios into a quarterly series ECONOMIC SHOCKS There were several major economic shocks during the period 1985 through 2009 that affected ferrous scrap supply. The full impact of these shocks might not be captured by variables we include in our model. Hence, we introduced binary (dummy) variables to measure impacts of these events on the supply of ferrous scrap. A dummy variable is set equal to one for the quarters the event might have affected supply. It is set equal to zero for all other quarters. Financial and Economic Crises in Asia and Brazil Between 1998 and 2002, imports of iron and steel scrap into the United States increased rapidly due mainly to low prices and low demand in Asia and Brazil in the aftermath of their financial and economic crises. Average U.S. imports of iron and steel scrap were 852,000 net tons per quarter during the period. In contrast, average imports were only 589,000 net tons during the prior five years. Hence, between 1998 and 2002, the abnormally high volume of imports likely had a disproportionate effect on scrap receipts vis-à-vis other periods. To account for this possibility a dummy variable was included in our model. Its value was set equal to one for each quarter of 1998 through 2002 and zero during all other quarters. 25 Note that the use of apparent steel consumption as a supply factor is an improvement to our 1979 model. In the earlier model, we used steel mill and foundry shipments. However, foundry shipments are of finished products such as tubes and castings that are not re-shaped. No prompt scrap is generated when foundry products are consumed. 26 See and steel statistics on pages five through eight of 27 For 1985 to 2003, exports are reported by USGS in its Open_file Report , Historical Statistics for Mineral and Material Commodities in the United States, Table: Iron and Steel Scrap Statistics. For , the data are reported in USGS Data Series 140, Historical Statistics for Mineral and Mineral Commodities in the United States, Table: Iron and Steel Scrap Statistics. A value for 2009 is reported in the Mineral Industry Survey of January 2010, Tables 7 and 9.

23 19 U.S. Financial Crisis and Economic Recession The steel and ferrous scrap industries were significantly affected by the severe U.S. recession from the fourth quarter of 2008 through The downturn in economic activity reduced scrap receipts and steel production. Hence, we included another dummy variable in our model. Its value equals one during the fourth quarter of 2008 and every quarter of 2009 and zero in all other quarters. 3.4 Factors Affecting Scrap Demand Other than Price Factors that might affect scrap demand other than the price of scrap include the production of crude steel, consumption of iron ore, technological change, and the availability of other scrap that can substitute for obsolete ferrous scrap CONSUMPTION OF IRON ORE Iron ore is the raw material used in crude steel production. In the form of pig iron, it substitutes for scrap. Hence, we include iron ore consumption at U.S plants 28 in our demand function to control for its impact on the demand for ferrous scrap CRUDE STEEL PRODUCTION Crude steel production drives the demand for ferrous scrap. Hence, we included production in our demand function. We summed monthly crude steel production volumes reported by the World Steel Association to obtain quarterly production data TECHNOLOGICAL CHANGE Technological changes in steelmaking, particularly the increased utilization of the basic oxygen furnace (BOF) and the electric arc furnace (EAF), affect scrap demand. Of the metal materials used by BOF to make crude steel, 35 percent to 40 percent are ferrous scrap and the remaining 65 percent to 60 percent are pig iron. In the case of EAF, metal material inputs are almost entirely ferrous scrap. Therefore, the demand for scrap is likely to be highly influenced by the proportion of steel produced in BOFs and EAFs. Hence, we included in our demand function the proportion of total steel produced by the BOF technology as a factor affecting demand. 30 However, we did not include the proportion produced by EAF to avoid the statistical problem of co-linearity between explanatory factors. The EAF proportion is highly correlated with the BOF proportion. We converted reported annual BOF proportions into quarterly values using linear interpolation. 28 Apparent consumption in metric tons reported by USGS at 29 See 30 Reported by USGS and available online at

24 AVAILABILITY OF OTHER SCRAP Finally, the quantity of purchased scrap demanded is at least partly influenced by the availability of home scrap. 31 Indeed, the quantity of obsolete scrap demanded is influenced by the availability of home and prompt scrap. Although home and prompt scrap supply are independent of price, they do affect market-clearing scrap prices. As additional volumes become available, the vertical supply curve (recall from Figure 3 above) shifts to the right. The shift pushes the total scrap supply curve to the right, which determines a new market-clearing scrap price-quantity combination. To control for these effects, production volumes of home scrap are included in our demand function as ratios. In one case, we include the ratio of home scrap to home plus purchased scrap. In another case, we include the ratio of home plus prompt scrap to home plus purchased (prompt and obsolete) scrap. While home scrap production figures are reported annually by the USGS, we estimated prompt scrap generated annually from shipments data, as discussed previously. Quarterly crude steel production figures were used to convert annual home scrap production into quarterly figures and data on quarterly apparent consumption (production plus imports minus exports) of iron and steel were used to convert estimated prompt scrap generated into a quarterly series Reported by USGS and available online at 32 Apparent consumption is calculated as shipments plus imports less exports of iron and steel. Monthly figures are reported by the American Iron and Steel Institute in its Annual Statistical Report, 2009, and were totaled to arrive at quarterly figures

25 4. Model Having identified factors affecting the supply and demand of ferrous scrap, we now turn to specifying a model. We specify two. The first model is of total purchased ferrous scrap, which consists of prompt and obsolete scrap. The second is of purchased obsolete scrap only. 4.1 Total Prompt and Obsolete Scrap For a reason that will soon become apparent, we begin by specifying the model in a multiplicative form as follows: 13. Total Q d t = A 0 (Price A t 1 ) (TSTL A t 2 ) (OREC A t 3 ) (BOP A t 4 ) (C A t 5 ) 14. Total Q S t = B 0 (Price B t 1 ) (COST B t 2 ) (FRTR B t 3 ) (APC B t 4 ) (XPRTR B t 5 ) (DUMMY B t 6 ) where Price is the producer price index for iron and steel scrap, TSTL is crude steel production, OREC is iron ore consumption, BOP is the BOF-produced share of total steel production, C is the home scrap production share of home and purchased scrap, COST is our index of ferrous scrap processing cost, FRTR is railroad revenue per ton-mile of waste and scrap, APC is apparent steel consumption, XPRTR is scrap exports as a share of net purchased scrap receipts, and DUMMY is a binary variable identifying quarters of economic shock. We next transform the model s demand and supply functions by taking the natural logarithm (ln) of each. For simplicity, we drop the subscript for time (t), which is quarterly in our data. 15. ln( Total Q d ) = A 0 + A 1 ln(price) + A 2 ln(tstl) + A 3 ln(orec) + A 4 ln(bop) + A 5 ln(c) 16. ln( Total Q S ) = B 0 + B 1 ln(price) + B 2 ln(cost) + B 3 ln(frtr) + B 4 ln(apc) + B 5 ln(xprtr) + B 6 ln(dummy) As expressed in equations 15 and 16 above, estimated values of A 1 and B 1 will be price demand and supply elasticities, respectively.

26 Obsolete Scrap We begin with the following specification (again multiplicative) for our model of obsolete scrap: 17. Obsolete Q d t = C 0 (Price C t 1 ) (TSTL C t 2 ) (OREC C t 3 ) (BOP C t 4 ) (B C t 5 ) 18. Obsolete Q S t = D 0 (Price D t 1 ) (COST D t 2 ) (FRTR D t 3 ) (XPRTR D t 4 ) (DUMMY D t 5 ) where all right-hand-side variables, but for B, are as defined above. In this model, B is home and prompt scrap as a share of home, prompt, and obsolete scrap. In the supply function, APC is not included. We are modeling purchased obsolete scrap, not total purchased scrap which includes prompt. The logarithmic transformation of the above functions yields the following: 19. ln( Obsolete Q d ) = C 0 + C 1 ln(price) + C 2 ln(tstl) + C 3 ln(orec) + C 4 ln(bop) + C 5 ln(b) 20. ln( Obsolete Q S ) = D 0 + D 1 ln(price) + D 2 ln(cost) + D 3 ln(frtr) + D 4 ln(xprtr) + D 5 ln(dummy) Again, as expressed in equations 19 and 20 above, estimated values of C 1 and D 1 will be price demand and supply elasticities, respectively. 4.3 Estimation A simple count of variables in each model reveals the supply and demand functions in each are over-specified. In both the total and the obsolete scrap models, m equals two. In the total scrap model, there are nine exogenous variables (four in the demand function and five in the supply function). In the demand function, the five exogenous variables of the supply function are missing, so k equals five, which is greater than m 1. Hence, the demand function is overspecified. In the supply function, k equals four, which exceeds m-1, so it too is over-specified. In the obsolete model, k equals four in both the demand and supply functions. Hence, both are over-specified (k > m-1). Hence, 2SLS is the appropriate technique for estimating the parameters of the models. In the first stage, OLS is used to estimate a proxy for the endogenous price variable. In the second stage, OLS is used again to estimate the supply and demand functions when the estimated values of the proxy for price are used instead of actual prices.

27 23 The specification for estimating a price proxy in the first stage of 2SLS is as follows: ln(price) = E 0 + E 1 ln(tstl) + E 2 ln(orec) + E 3 ln(bop) + E 4 ln(b) + E 5 ln(cost) + E 6 ln(frtr) + E 7 ln(xprtr) + E 8 ln(dummy The estimated parameters of equation 21 above, as well as statistical measures of the goodness of fit of the estimated equation are presented below (Table 2). Table 2 Regression Results for Estimation of Price Proxy (first stage of 2SLS) Parameter Variable Parameter Estimate Standard Error Calculated t-value /a Significance /b E0 Constant percent E1 TSTL percent E2 OREC > 10 percent E3 BOP > 10 percent E4 B percent E5 COST percent E6 FRTR percent E7 XPRTR > 10 percent E8 DUMMY percent Note: 100 observations were used in the regression. The adjusted R-square statistic is The F value of the regression is which is significant at one percent. a. The t-value is the estimated value of the coefficient divided by its standard error. It measures the estimated coefficient s number of standard deviations from a value of zero. It is used to test the hypothesis that the true value of the coefficient is non-zero. b. The probability of observing a t-statistic as large or larger in magnitude as the calculated t- value given that the true value of the coefficient is zero. A significance level of 5 percent indicates a 5 percent chance of rejecting the null hypothesis (true value of the coefficient is zero) when it is true. Once the proxy function for price has been estimated, estimated quarterly values of Price t for 1985 through 2009 can be calculated from the actual values of the exogenous variables and the estimated coefficients of the equation. Estimated quarterly values of Price t are then used in the second stage of 2SLS to estimate the parameters of the supply and demand functions in both the total purchased scrap model and the obsolete scrap model. 33 Notice that ln(c) is not included. Both ln(c) and ln(b) cannot be included because they are not independent. Home scrap as a share of home and purchased scrap (C) is correlated with home and prompt scrap as a share of home and purchased scrap (B).